This is an automated email from the ASF dual-hosted git repository.
morningman pushed a commit to branch branch-4.0
in repository https://gitbox.apache.org/repos/asf/doris.git
The following commit(s) were added to refs/heads/branch-4.0 by this push:
new 28add521d04 branch-4.0: [fix](test) deflake test_curd_wlg and adapt
test_single_replica_compaction_with_format_v2 to V3 default (#64666)
28add521d04 is described below
commit 28add521d043de755c6d705b5972d731228c6a5c
Author: Mingyu Chen (Rayner) <[email protected]>
AuthorDate: Mon Jun 22 16:07:53 2026 +0800
branch-4.0: [fix](test) deflake test_curd_wlg and adapt
test_single_replica_compaction_with_format_v2 to V3 default (#64666)
## Proposed changes
Fix two pre-existing branch-4.0 P0 regression-test failures (surfaced by
an internal P0 build). Both are test-side issues, unrelated to product
code.
### 1. `test_single_replica_compaction_with_format_v2` — failed 100% on
master/4.1/4.0, passes on 3.0
`#57252` changed the FE default `inverted_index_storage_format` from
`V2` to `V3`, but this test still created its table without pinning the
property and asserted `check_nested_index_file(..., 2, 3, "V2")` with a
hardcoded rowset count. Its sibling `*_format_v2` tests
(`test_cumulative_compaction_with_format_v2`,
`test_mow_table_with_format_v2`) were already migrated to a dynamic
rowset count + `"V3"`; this one was missed.
Fix: derive the post-compaction rowset count per tablet via
`be_show_tablet_status` and assert `"V3"`, mirroring the sibling tests.
### 2. `test_curd_wlg` — flaky
The `bypass_group` (`max_concurrency=0`) check expects `select count(1)
from wlg_test_table` to be rejected with `query waiting queue is full`.
That identical query is run repeatedly earlier in the suite, so it can
be served from the FE SQL result cache, which returns before a
Coordinator is built and never enters the query queue — so the rejection
never fires and the assertion sees no exception (`but meet 'null'`).
Fix: disable the SQL cache (`set enable_sql_cache=false`) around this
assertion so the query actually executes and hits the queue, then
restore it.
## Further comments
Test-only changes. Cannot be verified locally (needs a deployed cluster
with >= 2 BEs); please verify via CI.
---
.../test_single_replica_compaction_with_format_v2.groovy | 7 ++++++-
regression-test/suites/workload_manager_p0/test_curd_wlg.groovy | 5 +++++
2 files changed, 11 insertions(+), 1 deletion(-)
diff --git
a/regression-test/suites/inverted_index_p0/index_format_v2/test_single_replica_compaction_with_format_v2.groovy
b/regression-test/suites/inverted_index_p0/index_format_v2/test_single_replica_compaction_with_format_v2.groovy
index 2d267cf31b2..b0f87cd8bf6 100644
---
a/regression-test/suites/inverted_index_p0/index_format_v2/test_single_replica_compaction_with_format_v2.groovy
+++
b/regression-test/suites/inverted_index_p0/index_format_v2/test_single_replica_compaction_with_format_v2.groovy
@@ -183,7 +183,12 @@ suite("test_single_replica_compaction_with_format_v2",
"inverted_index_format_v2
backend_id = tablet.BackendId
String ip = backendId_to_backendIP.get(backend_id)
String port = backendId_to_backendHttpPort.get(backend_id)
- check_nested_index_file(ip, port, tablet_id, 2, 3, "V2")
+ // The default inverted_index_storage_format changed from V2 to V3
(#57252), and the
+ // post-compaction rowset count varies per replica, so derive it
dynamically and check
+ // against V3 (mirrors test_cumulative_compaction_with_format_v2 /
test_mow_table_with_format_v2).
+ def (tablet_status_code, tablet_status_out, tablet_status_err) =
be_show_tablet_status(ip, port, tablet_id)
+ int activeRowsetCount =
parseJson(tablet_status_out.trim()).rowsets.size()
+ check_nested_index_file(ip, port, tablet_id, activeRowsetCount, 3,
"V3")
}
int segmentsCount = 0
diff --git a/regression-test/suites/workload_manager_p0/test_curd_wlg.groovy
b/regression-test/suites/workload_manager_p0/test_curd_wlg.groovy
index c32c5676330..b560644975b 100644
--- a/regression-test/suites/workload_manager_p0/test_curd_wlg.groovy
+++ b/regression-test/suites/workload_manager_p0/test_curd_wlg.groovy
@@ -417,10 +417,15 @@ suite("test_crud_wlg") {
// test bypass
sql "create workload group if not exists bypass_group $forComputeGroupStr
properties ( 'max_concurrency'='0','max_queue_size'='0','queue_timeout'='0');"
sql "set workload_group=bypass_group;"
+ // Disable the FE SQL cache here: `select count(1) from ${table_name}` is
run repeatedly in this
+ // suite, so it can be served from the result cache and skip the
Coordinator/query-queue path
+ // entirely, which makes this "queue is full" assertion flaky (the
rejection never fires).
+ sql "set enable_sql_cache=false;"
test {
sql "select count(1) from ${table_name};"
exception "query waiting queue is full"
}
+ sql "set enable_sql_cache=true;"
sql "set bypass_workload_group = true;"
sql "select count(1) from information_schema.active_queries;"
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
